• 인터넷정보학회논문지
• /
• 제20권4호
• /
• pp.91-102
• /
• 2019

자율주행 시스템에서, 카메라에 포착된 영상을 통하여 보행자를 분류하는 기능은 보행자 안전을 위하여 매우 중요하다. 기존에는 HOG(Histogram of Oriented Gradients)나 SIFT(Scale-Invariant Feature Transform) 등으로 보행자의 특징을 추출한 후 SVM(Support Vector Machine)으로 분류하는 기술을 사용했었으나, 보행자 특징을 위와 같이 수동(handcrafted)으로 추출하는 것은 많은 한계점을 가지고 있다. 따라서 본 논문에서는 CNN(Convolutional Neural Network)의 깊은 특징(deep features)과 전이학습(transfer learning)을 사용하여 보행자를 안정적이고 효과적으로 분류하는 방법을 제시한다. 본 논문은 2가지 대표적인 전이학습 기법인 고정특징추출(fixed feature extractor) 기법과 미세조정(fine-tuning) 기법을 모두 사용하여 실험하였고, 특히 미세조정 기법에서는 3가지 다른 크기로 레이어를 전이구간과 비전이구간으로 구분한 후, 비전이구간에 속한 레이어들에 대해서만 가중치를 조정하는 설정(M-Fine: Modified Fine-tuning)을 새롭게 추가하였다. 5가지 CNN모델(VGGNet, DenseNet, Inception V3, Xception, MobileNet)과 INRIA Person데이터 세트로 실험한 결과, HOG나 SIFT 같은 수동적인 특징보다 CNN의 깊은 특징이 더 좋은 성능을 보여주었고, Xception의 정확도(임계치 = 0.5)가 99.61%로 가장 높았다. Xception과 유사한 성능을 내면서도 80% 적은 파라메터를 학습한 MobileNet이 효율성 측면에서는 가장 뛰어났다. 그리고 3가지 전이학습 기법중 미세조정 기법의 성능이 가장 우수하였고, M-Fine 기법의 성능은 미세조정 기법과 대등하거나 조금 낮았지만 고정특징추출 기법보다는 높았다.

• 한국멀티미디어학회논문지
• /
• 제23권8호
• /
• pp.927-939
• /
• 2020

Gait recognition can identify people's identity from a long distance, which is very important for improving the intelligence of the monitoring system. Among many human features, gait features have the advantages of being remotely available, robust, and secure. Traditional gait feature extraction, affected by the development of behavior recognition, can only rely on manual feature extraction, which cannot meet the needs of fine gait recognition. The emergence of deep convolutional neural networks has made researchers get rid of complex feature design engineering, and can automatically learn available features through data, which has been widely used. In this paper,conduct feature metric learning in the three-dimensional space by combining the three-dimensional convolution features of the gait sequence and the Siamese structure. This method can capture the information of spatial dimension and time dimension from the continuous periodic gait sequence, and further improve the accuracy and practicability of gait recognition.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제17권8호
• /
• pp.2053-2067
• /
• 2023

This paper proposes two video quality assessment methods based on deep neural network. (i)The first method uses the IQF-CNN (convolution neural network based on image quality features) to build image quality assessment method. The LIVE image database is used to test this method, the experiment show that it is effective. Therefore, this method is extended to the video quality assessment. At first every image frame of video is predicted, next the relationship between different image frames are analyzed by the hysteresis function and different window function to improve the accuracy of video quality assessment. (ii)The second method proposes a video quality assessment method based on convolution neural network (CNN) and gated circular unit network (GRU). First, the spatial features of video frames are extracted using CNN network, next the temporal features of the video frame using GRU network. Finally the extracted temporal and spatial features are analyzed by full connection layer of CNN network to obtain the video quality assessment score. All the above proposed methods are verified on the video databases, and compared with other methods.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제17권3호
• /
• pp.721-739
• /
• 2023

Rating prediction is an important issue in recommender systems, and its accuracy affects the experience of the user and the revenue of the company. Traditional recommender systems use Factorization Machinesfor rating predictions and each feature is selected with the same weight. Thus, there are problems with inaccurate ratings and limited data representation. This study proposes a deep recommendation model based on self-attention Factorization (SAFMR) to solve these problems. This model uses Convolutional Neural Networks to extract features from user and item reviews. The obtained features are fed into self-attention mechanism Factorization Machines, where the self-attention network automatically learns the dependencies of the features and distinguishes the weights of the different features, thereby reducing the prediction error. The model was experimentally evaluated using six classes of dataset. We compared MSE, NDCG and time for several real datasets. The experiment demonstrated that the SAFMR model achieved excellent rating prediction results and recommendation correlations, thereby verifying the effectiveness of the model.

• 한국IT서비스학회지
• /
• 제18권2호
• /
• pp.143-159
• /
• 2019

Machine learning (ML) is a method of fitting given data to a mathematical model to derive insights or to predict. In the age of big data, where the amount of available data increases exponentially due to the development of information technology and smart devices, ML shows high prediction performance due to pattern detection without bias. The feature engineering that generates the features that can explain the problem to be solved in the ML process has a great influence on the performance and its importance is continuously emphasized. Despite this importance, however, it is still considered a difficult task as it requires a thorough understanding of the domain characteristics as well as an understanding of source data and the iterative procedure. Therefore, we propose methods to apply deep learning for solving the complexity and difficulty of feature extraction and improving the performance of ML model. Unlike other techniques, the most common reason for the superior performance of deep learning techniques in complex unstructured data processing is that it is possible to extract features from the source data itself. In order to apply these advantages to the business problems, we propose deep learning based methods that can automatically extract features from transaction data or directly predict and classify target variables. In particular, we applied techniques that show high performance in existing text processing based on the structural similarity between transaction data and text data. And we also verified the suitability of each method according to the characteristics of transaction data. Through our study, it is possible not only to search for the possibility of automated feature extraction but also to obtain a benchmark model that shows a certain level of performance before performing the feature extraction task by a human. In addition, it is expected that it will be able to provide guidelines for choosing a suitable deep learning model based on the business problem and the data characteristics.

• International Journal of Computer Science & Network Security
• /
• 제21권7호
• /
• pp.159-168
• /
• 2021

Detecting cyber-attacks using machine learning or deep learning is being studied and applied widely in network intrusion detection systems. We noticed that the application of deep learning algorithms yielded many good results. However, because each deep learning model has different architecture and characteristics with certain advantages and disadvantages, so those deep learning models are only suitable for specific datasets or features. In this paper, in order to optimize the process of detecting cyber-attacks, we propose the idea of building a new deep learning network model based on the association and combination of individual deep learning models. In particular, based on the architecture of 2 deep learning models: Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM), we combine them into a combined deep learning network for detecting cyber-attacks based on network traffic. The experimental results in Section IV.D have demonstrated that our proposal using the CNN-LSTM deep learning model for detecting cyber-attacks based on network traffic is completely correct because the results of this model are much better than some individual deep learning models on all measures.

• Journal of Information Processing Systems
• /
• 제17권2호
• /
• pp.337-351
• /
• 2021

Automatically recognizing facial expressions in video sequences is a challenging task because there is little direct correlation between facial features and subjective emotions in video. To overcome the problem, a video facial expression recognition method using spatiotemporal recurrent neural network and feature fusion is proposed. Firstly, the video is preprocessed. Then, the double-layer cascade structure is used to detect a face in a video image. In addition, two deep convolutional neural networks are used to extract the time-domain and airspace facial features in the video. The spatial convolutional neural network is used to extract the spatial information features from each frame of the static expression images in the video. The temporal convolutional neural network is used to extract the dynamic information features from the optical flow information from multiple frames of expression images in the video. A multiplication fusion is performed with the spatiotemporal features learned by the two deep convolutional neural networks. Finally, the fused features are input to the support vector machine to realize the facial expression classification task. The experimental results on cNTERFACE, RML, and AFEW6.0 datasets show that the recognition rates obtained by the proposed method are as high as 88.67%, 70.32%, and 63.84%, respectively. Comparative experiments show that the proposed method obtains higher recognition accuracy than other recently reported methods.

• 대한원격탐사학회지
• /
• 제37권4호
• /
• pp.719-731
• /
• 2021

This study proposes a two-stage hybrid classification model for crop classification using multi-temporal remote sensing images; the model combines feature embedding by using an autoencoder (AE) with a convolutional neural network (CNN) classifier to fully utilize features including informative temporal and spatial signatures. Long short-term memory (LSTM)-based AE (LAE) is fine-tuned using class label information to extract latent features that contain less noise and useful temporal signatures. The CNN classifier is then applied to effectively account for the spatial characteristics of the extracted latent features. A crop classification experiment with multi-temporal unmanned aerial vehicle images is conducted to illustrate the potential application of the proposed hybrid model. The classification performance of the proposed model is compared with various combinations of conventional deep learning models (CNN, LSTM, and convolutional LSTM) and different inputs (original multi-temporal images and features from stacked AE). From the crop classification experiment, the best classification accuracy was achieved by the proposed model that utilized the latent features by fine-tuned LAE as input for the CNN classifier. The latent features that contain useful temporal signatures and are less noisy could increase the class separability between crops with similar spectral signatures, thereby leading to superior classification accuracy. The experimental results demonstrate the importance of effective feature extraction and the potential of the proposed classification model for crop classification using multi-temporal remote sensing images.

• 방송공학회논문지
• /
• 제26권7호
• /
• pp.868-876
• /
• 2021

Deploying deep neural network model training performs remarkable performance in the fields of Object detection and Instance segmentation. To train these models, features are first extracted from the input image using a backbone network. The extracted features can be reused by various tasks. Research has been actively conducted to serve various tasks by using these learned features. In this process, standardization discussions about encoding, decoding, and transmission methods are proceeding actively. In this scenario, it is necessary to analyze the response characteristics of features against various distortions that may occur in the data transmission or data compression process. In this paper, experiment was conducted to inject various distortions into the feature in the object recognition task. And analyze the mAP (mean Average Precision) metric between the predicted value output from the neural network and the target value as the intensity of various distortions was increased. Experiments have shown that features are more robust to distortion than images. And this points out that using the feature as transmission means can prevent the loss of information against the various distortions during data transmission and compression process.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권1호
• /
• pp.105-125
• /
• 2024

Gliomas are the most common malignant brain tumor and cause the most deaths. Manual brain tumor segmentation is expensive, time-consuming, error-prone, and dependent on the radiologist's expertise and experience. Manual brain tumor segmentation outcomes by different radiologists for the same patient may differ. Thus, more robust, and dependable methods are needed. Medical imaging researchers produced numerous semi-automatic and fully automatic brain tumor segmentation algorithms using ML pipelines and accurate (handcrafted feature-based, etc.) or data-driven strategies. Current methods use CNN or handmade features such symmetry analysis, alignment-based features analysis, or textural qualities. CNN approaches provide unsupervised features, while manual features model domain knowledge. Cascaded algorithms may outperform feature-based or data-driven like CNN methods. A revolutionary cascaded strategy is presented that intelligently supplies CNN with past information from handmade feature-based ML algorithms. Each patient receives manual ground truth and four MRI modalities (T1, T1c, T2, and FLAIR). Handcrafted characteristics and deep learning are used to segment brain tumors in a Global Convolutional Neural Network (GCNN). The proposed GCNN architecture with two parallel CNNs, CSPathways CNN (CSPCNN) and MRI Pathways CNN (MRIPCNN), segmented BraTS brain tumors with high accuracy. The proposed model achieved a Dice score of 87% higher than the state of the art. This research could improve brain tumor segmentation, helping clinicians diagnose and treat patients.